Differences between clang and gcc handling of int[static n] function arguments

[peter0x44 <peter0x44@disroot.org>]

[-- Attachment #1: Type: text/plain, Size: 2992 bytes --]



Hi,

So, recently I learned about the c99 feature to get NULL pointer checks 
for array function arguments.
I have really never seen this feature used in an actual codebase. It's 
definitely something I wanted on a few occasions.

To be clear, I'm talking about specifically:
void foo(int array[static 1]);

I checked what warnings this produces - gcc by default produces, none, 
but with -Wall it produces for this code:

int foo(int array[static 1]){ return array[0]; }

int main(void)
{
#define NULL (void*)0
        foo(NULL);
}

bruh.c: In function 'main':
bruh.c:8:9: warning: argument 1 to 'int[static 1]' is null where 
non-null expected [-Wnonnull]
    8 |         foo(NULL);
      |         ^~~~~~~~~
bruh.c:3:5: note: in a call to function 'foo'
    3 | int foo(int array[static 1]){ return array[0]; }
      |     ^~~
I think this warning is acceptable, but has some scope for improvement.

I checked what clang did instead, and it seemed nicer, for sure.

bruh.c:8:2: warning: null passed to a callee that requires a non-null 
argument [-Wnonnull]
        foo(NULL);
         ^   ~~~~
bruh.c:3:13: note: callee declares array parameter as static here
int foo(int array[static 1]){ return array[0]; }
             ^    ~~~~~~~~~~
  It's pointing me exactly to the parameter with the static directly, so 
there is no ambiguity

Also, this is a warning enabled by default, no need to pass -Wall.

Is there a reason gcc doesn't enable this by default? To me, it seems 
like a warning that's desirable always.

You are explicitly agreeing to never call these functions with NULL, any 
code doing that is surely broken.

There's no way this gives a false positive, ever.

I'm definitely adding this warning to -Werror on all of my future 
projects, now that I know about it.

One last thing worth mentioning, is that GCC makes a nicer warning than 
clang when this is done through __attribute__((nonnull))

bruh.c: In function 'main':
bruh.c:8:9: warning: argument 1 null where non-null expected [-Wnonnull]
    8 |         foo(NULL);
      |         ^~~
bruh.c:3:5: note: in a call to function 'foo' declared 'nonnull'
    3 | int foo(int array[1]){ return array[0]; }
      |     ^~~
  It points out specifically that that it is done through the attribute.

I think it would be nice if the attribute could be underlined also, 
though.

Clang produces:

bruh.c:8:10: warning: null passed to a callee that requires a non-null 
argument [-Wnonnull]
        foo(NULL);
             ~~~~^
  with no mention that the warning is specifically because of the 
attribute.

I tried looking on the bug tracker and I could find nothing elaborating 
on this. Maybe I'm not looking hard enough.

I would be happy to open a PR to improve this warning, if there isn't 
one already.

It seems it might even be trivial enough for me to investigate and 
tackle myself, in some spare time.

I see very little code using either of these features, so it's 
definitely not a high priority task regardless.